Introduction
This work introduces the first wireless embedded system capable of "fast" feedback control—with update intervals between 20 and 50 ms—over multi-hop networks. By co-designing a jitter-free communication stack with a control strategy that accounts for message loss and dynamic mode changes, the system provides provable closed-loop stability guarantees for demanding mechanical tasks, such as stabilizing inverted pendulums across a building.
The Performance Gap in Cyber-Physical Systems
Cyber-Physical Systems (CPS) often face a trade-off between speed and range. Controlling fast-moving robots or drones requires millisecond-level updates. Historically, researchers had to choose between fast single-hop links (limited range) or long-range multi-hop networks (slow updates).
Multi-hop networks typically introduce latency, jitter, and correlated message loss. Furthermore, real-world applications require "mode changes" (e.g., switching from takeoff to formation flight). Ensuring stability during these transitions over unreliable wireless links is a significant challenge.
Co-Design Methodology
The authors bridge this gap through a rigorous co-design of hardware, network protocols, and control theory.
1. Wireless Embedded System
The system utilizes a Dual-Processor Platform (DPP) where a Communication Processor (CP) manages the radio and an Application Processor (AP) handles control logic. They are connected via "Bolt," a predictable interconnect that isolates timing requirements. The network uses Glossy, a protocol based on synchronous transmissions that provides microsecond-level synchronization and topology independence.
2. Control Design and Stability
To handle the inherent one-step delay of wireless communication, the controller uses state prediction based on a mathematical model. The system is modeled as a stochastic switched linear system. Using Linear Matrix Inequalities (LMIs), the authors prove "mean-square stability," ensuring the system remains stable despite message loss. For mode changes, they calculate an "Average Dwell Time" to ensure stability during transitions.
Experimental Evaluation
The system was tested on a 20-node wireless testbed in an office building (15m x 20m) with a 3-hop diameter. The physical systems included two real "cart-pole" inverted pendulums and three simulated ones. Inverted pendulums serve as a rigorous benchmark due to their inherent instability.
Key Insights
- Jitter Reduction: By combining synchronous transmissions with the dual-processor architecture, timing jitter was reduced to ±50 microseconds, allowing the control theory to treat the network as nearly perfectly periodic.
- Topology Independence: Flooding-based protocols make the system robust to environmental changes or node movement.
- Predictability by Design: Building a predictable network simplifies the control design, making it more robust than trying to compensate for a chaotic network in software.
Results
- High-Speed Stability: Successfully stabilized real inverted pendulums over 3 hops with update intervals as low as 20 ms.
- Resilience: Maintained stability despite 75% intentional message loss or bursts of 40 consecutive losses.
- Seamless Transitions: Demonstrated stable runtime switching between different control tasks.
- Efficiency: Achieved these results with a low radio duty cycle (~15%), suitable for battery-powered operation.
Limitations and Future Work
The system currently supports up to five agents; scalability is limited as update intervals increase with network size. Future work could leverage higher bitrate radios (like BLE 5) to support more agents or faster updates. Additionally, while the current stability proofs focus on linear systems, extending this to non-linear dynamics remains an open challenge.
Summary updated